The Study Of Melody Extractionbased

Melody is the fundamental frequency of human voice in music.Melody extraction is defined as transcription melody of a piece of music to digital signal.Melody extraction is valuable in many music research fields.In this paper,the main melody extraction method is composed of two steps:the multi pitch extraction and the main melody decision.The main work is divided into the following two parts:(1)Studied two kinds of multi-pitch extraction algorithms and tried to make further improvementIn the pitch extraction part,we apply an equal loudness filter on the song signal to weaken the high frequency and low frequency part of signal energy song.After analyzing the shortage of Short-time Fourier Transform(STFT),we introduce Multi-Resolution Fast Fourier Transform(MRFFT)which use different window size to computing the spectrum of a frame.Then we introduce the method of sub harmonic superposition in the extraction of the muti-pitch extraction,and analyze the problems this method.Then we described the original Harmonic Pitch-Class Profile(HPCP)algorithm,and improved it to the muti-pitch extraction.Experimental results show that the new method is more accurate and simple.(2)Studied the melody extraction algorithms and proposed a melody decision algorithm which is combined with the rules and statistical methods.In the part of melody extraction,we determine which candidate pitch is the main melody.We realize the existing methods of candidate pitch into pitch contours.By studying the feature distributions of melodic and non-melodic contours we able to formulate filtering rules and train the support vector machine(SVM)model.By distinguishing between the melodic and non-melodic contours,finally extract the melody information.